+ + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +

arm_status arm_depthwise_conv_u8_basic_ver1	(	const uint8_t *	input,
		const uint16_t	input_x,
		const uint16_t	input_y,
		const uint16_t	input_ch,
		const uint8_t *	kernel,
		const uint16_t	kernel_x,
		const uint16_t	kernel_y,
		const int16_t	ch_mult,
		const int16_t	pad_x,
		const int16_t	pad_y,
		const int16_t	stride_x,
		const int16_t	stride_y,
		const int16_t	dilation_x,
		const int16_t	dilation_y,
		const int32_t *	bias,
		const int32_t	input_offset,
		const int32_t	filter_offset,
		const int32_t	output_offset,
		uint8_t *	output,
		const uint16_t	output_x,
		const uint16_t	output_y,
		const int32_t	output_activation_min,
		const int32_t	output_activation_max,
		const int32_t	out_shift,
		const int32_t	out_mult
	)

uint8 depthwise convolution function with asymmetric quantization for even number of channel multiplier and input channels. Unless specified otherwise, arguments are mandatory.

Parameters

+ + + + + + + + + + + + + + + + + + + + + + + + + + +

[in]	input	Pointer to input tensor
[in]	input_x	Width of input tensor
[in]	input_y	Height of input tensor
[in]	input_ch	Channels in input tensor
[in]	kernel	Pointer to kernel weights
[in]	kernel_x	Width of kernel
[in]	kernel_y	Height of kernel
[in]	ch_mult	Number of channel multiplier
[in]	pad_x	Padding sizes x
[in]	pad_y	Padding sizes y
[in]	stride_x	Convolution stride along the width
[in]	stride_y	Convolution stride along the height
[in]	dilation_x	Dilation along width. Not used and intended for future enhancement.
[in]	dilation_y	Dilation along height. Not used and intended for future enhancement.
[in]	bias	Pointer to optional bias values. If no bias is availble, NULL is expected
[in]	input_offset	Input tensor zero offset
[in]	filter_offset	Kernel tensor zero offset
[in]	output_offset	Output tensor zero offset
[in,out]	output	Pointer to output tensor
[in]	output_x	Width of output tensor
[in]	output_y	Height of output tensor
[in]	output_activation_min	Minimum value to clamp the output to. Range : {0, 255}
[in]	output_activation_max	Minimum value to clamp the output to. Range : {0, 255}
[in]	out_shift	Amount of right-shift for output
[in]	out_mult	Output multiplier for requantization

Returns: The function returns one of the following ARM_MATH_SIZE_MISMATCH - Not supported dimension of tensors ARM_MATH_SUCCESS - Successful operation ARM_MATH_ARGUMENT_ERROR - Implementation not available

Input constraints ch_mult is multiple of 2 kernel_x is multiple of 2

+ +

References arm_nn_divide_by_power_of_two(), arm_nn_sat_doubling_high_mult(), DILATION_X, DILATION_Y, LEFT_SHIFT, and RIGHT_SHIFT.